Search the archive

THE AI GRAVEYARD: A RUNNING LIST OF PROJECTS AND STARTUPS THAT DIDN'T MAKE IT

Relay, an AI workflow tool, shut down due to competition from larger companies integrating similar features.

PROGRAMMING IS DEAD. ENGINEERING IS THRIVING

G5 Labs, an MIT CSAIL spinout, raised $14 million to develop a platform where natural language, structured as intent graphs, serves as the source code.

GLM 5.3 and GLM 5.3 Flash are now in Crusoe Intelligence Foundry.

Sign up and let these powerful models handle your most difficult software engineering workflows, [6]at a small fraction of the cost of proprietary offers - but with the same serverless, frictionless experience. ➡️ GLM-5.3: Z.ai's flagship open-weights AI model released on August 14, 2026, built specifically for advanced software engineering and autonomous agent tasks. ✅ Achieves open-source SOTA on Terminal Bench 3.0 and Agents' Last Exam ➡️ GLM-5.3-FLASH: MIT licensed with weights already released. Roughly 9x cheaper for workloads that do not need the frontier tier.

Create your free account and start building

APPLE'S SIRI AI CAN BE SWAPPED OUT FOR CLAUDE, CHATGPT, CODE SHOWS

Private iOS 27 and macOS frameworks show Apple built Siri so Claude or GPT-5.6 can sit in as the brain: Model Delegation lets Claude handle a natural-language ask, then hand Reminders or Messages back to Siri, and an inference-provider path can swap Apple's server-side Siri model entirely while keeping Siri's UI and voice. A demo already has ChatGPT finding emails, summarizing action items, and texting a contact through Siri's tools. The hooks are not user-facing yet, but they land as iOS 27 ships and after EU DMA pressure to open Siri to rivals.

ANTHROPIC PREPARES CLAUDE MONEY FOR PERSONAL FINANCE

Claude Money allows users to link their bank accounts and ask Claude about spending, plans, and other financial questions. It appears to be designed to give Claude persistent access to users' financial context, removing the need to manually upload bank statements or transaction exports. The exact account types, data provider, and supported actions remain unknown. Claude already supports third-party personal-finance connectors, and Anthropic has been building finance-specific agents and products for professional users. The launch timeframe remains unknown.

OPENAI BUYS STARTUP DEVELOPING SMARTPHONE CAMERA

OpenAI has quietly bought Glass Imaging, a company developing smartphone cameras that use AI to produce image quality on par with a DSLR camera. The deal values Glass Imaging at over $300 million. OpenAI's plans for the company are unclear, but it has been developing a secretive device with Jony Ive, the former Apple executive who helped design the iPhone. Glass Imaging was founded in 2019 by two former Apple employees who had previously worked on camera technology at the company.

WHAT DOES PACING MEAN?

Dario Amodei called for the AI industry to self-regulate its pace, but five camps interpreted "pacing" differently without agreeing on a speed. These camps focus on model interpretability, worker interests, economic growth, geopolitical strategy, and resisting new regulations. A previous attempt to limit AI training compute failed, highlighting the challenge of controlling AI innovation speed.

BAD BENCHMARKS AND EVALS: SENIOR SWE-BENCH, NAPKIN MATH, AND WINTER TIRES

Evals are more about avoiding mistakes than following some particular process. There isn't any step-by-step guide that will work, but there are approaches to experimental design that can teach you how to do better. This post aims to help people become better at benchmarking and evals. It looks at three different kinds of benchmarks: one for baseline numbers for performance 'napkin math' estimates, one set of AI model evals, and one on car tires.

WHO GETS TO DEFINE THE RULES FOR AI?

Aidan Gomez, Cohere CEO, argues against a few dominant Silicon Valley firms setting global AI regulations, drawing parallels to historical monopolies that limited competition under safety pretexts. He proposes a diverse, international, evidence-based framework for AI governance, emphasizing transparency, mandatory testing, and independent assurance mechanisms. Gomez warns that current proposals could entrench market leaders, advocating for more inclusive rule-making to foster competition and trust.

days/hours

AUGMENTED LAGRANGIAN PREDICTIVE CODING

PC-ALM is a local alternative to backpropagation. It is an extension of standard predictive coding (PC), which uses diffusive coupling between layers. PC-ALM trains residual MLPs up to 1,000 layers, nearly matching backprop's performance despite using only layer-local dynamics. It equips each layer with a feedback control dynamical system that distributes and propagates supervision credit throughout a network.

TAU

Tau is a coding agent that lives in the terminal. It can read files, edit code, run commands, and keep a durable session history while streaming what it is doing. Tau was designed to be a teaching project for understanding the shape of a coding-agent system without starting from a giant production codebase. It ships with support for OpenAI-compatible endpoints.

ARTEMIS

ARTEMIS lets AI assistants and test suites use real phones like a human. It can execute testing workflows and everyday tasks on Android from natural language instructions, using element indices when available, with coordinate and visual locating fallbacks for custom interfaces. ARTEMIS checks targets before individual actions and returns blocked actions to the Operator for recovery. It supports long-running exploratory and stability tests.

STEPAUDIO 3 TECHNICAL REPORT

StepAudio 3 Gen uses discrete autoregressive modeling over shared RVQ audio tokens to generate speech, voices, vocals, sound effects, music, and mixed audio within one model. It achieved state-of-the-art results on text-to-speech and voice design while retaining broad general-audio generation capabilities.

SAM ALTMAN BACKS FEDERAL FRONTIER AI SAFETY RULES

Sam Altman supports consistent federal safety requirements for frontier AI while urging labs to act before legislation arrives. OpenAI now uses safety cases before major reinforcement learning runs, arguing that capability growth should be deliberately paced alongside alignment and monitoring.

PERSONAL STATEMENT ON AI RISK

Models are becoming so situationally aware that researchers are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled. Models increasingly seem aligned even when they are not. The current evidence suggests that the field will continue to make progress at a fast pace. Once language models reach the capability threshold where they can shape the world unconstrained by human will, they might do something extreme and destroy humanity in the process.

FRONTIER LABS HAVE A FINANCIAL INCENTIVE TO PACE THE FRONTIER

Frontier labs are arguing for regulatory pacing on safety grounds. While many are undoubtedly well-intentioned, it is also true that the rules the labs have proposed protect their investments, preserve price premiums, and defer billions in competitive spending. Their arguments for these rules should not be taken at face value. The labs can not economically slow down on their own without the government stepping in to enforce a slowdown.

WHY WE BUILT PION

Pion is an agent designed to run any company fully autonomously.

OPENAI'S NEXT CHATGPT AD FORMAT: CLICK TO CHAT, NOT TO SITE

OpenAI is rolling out a new AI-native ChatGPT ad to select clients that opens a branded agent chat instead of a website.

CLINE DESKTOP: AN OPEN-SOURCE APP FOR OPEN-WEIGHT MODELS

Cline Desktop is an open-source app for open-weight models that can run parallel AI agents and automate recurring tasks.

ARTIFICIAL ANALYSIS CAPABILITY INDICES V1.1

The Artificial Analysis Capability Indices have been updated with stronger domain tuning, with Claude Fable 5.1 (max) leading in all six indices.

TURNBENCH: A MULTI-DOMAIN BENCHMARK FOR TURN-TAKING DYNAMICS IN SPOKEN DIALOGUE

TURNBENCH introduces a 30-hour, six-domain benchmark for end-of-turn and interruption detection, plus a 104-hour training set.

PERPLEXITY PORTABLE COMPUTER IS NOW AVAILABLE ON WINDOWS, POWERED BY NVIDIA RTX

NVIDIA's Perplexity Portable Computer is now available on Windows RTX PCs, enabling local AI tasks through powerful agentic AI.

The result: a reproducible framework that pushed MFU past 60%

(25%+ improvement vs. industry benchmark) with no changes to model architecture.

Get the whitepaper.

WE MUST PACE THE FRONTIER

Anthropic CEO Dario Amodei called for slowing the rate of frontier AI capability development and proposed measures including independent evaluators to verify safety commitments and incident reporting.

INTRODUCING PROJECTS

Cursor Projects lets users take on larger bodies of work. It can maintain context over months of work, delegate tasks to thousands of agents, and perform recurring work without being prompted. Cursor Projects frees developers from managing agents and lets them direct the work itself by moving up a level of abstraction. The tool has been a substantial productivity multiplier at Cursor: new users merge 30% more PRs.

ARC-AGI-4

ARC Prize believes that open source will be the foundation for advanced AI capable of scientific innovation. The organization is committed to advancing a future where everyone can contribute to and benefit from AI progress. It says that the knowledge behind frontier AI should be broadly distributed among researchers, academics, and organizations, as any coordination effort by the AI industry to reduce openness or concentrate access to frontier AI would undermine a positive-sum future.

OPENAI PUSHES ITS IPO BEYOND 2026

Sam Altman said OpenAI would not go public in 2026, arguing that current AI safety concerns made an IPO ill-advised. The company had previously filed confidentially and was reportedly considering a 2027 listing instead.

SOFTBANK GETS UPSIZED $11.9 BILLION LOAN IN OPENAI FUNDING PUSH

SoftBank just borrowed nearly $12 billion from about 20 banks to keep funding OpenAI, beating the $10 billion it first sought. Son is still aiming near $65 billion into OpenAI by October even as Altman shelves a 2026 IPO over safety and SoftBank shares sank as much as 13% Monday on the debt pile.

AI RESEARCHERS DEBATE HOW CLOSE WE ARE TO RECURSIVE SELF-IMPROVEMENT

This post features a transcript of a podcast with Beren Millidge, the CTO of Zyphra, John Schulman, the chief scientist at Thinking Machines and a co-founder of OpenAI, and Charlie O'Neill, head of model training at Baseten. The episode uncovers the details of what's happening at the frontier and what comes next. A link to the full video is available.

GPT-6-ASTRA CAN DO AMBITIOUS THINGS

Astra likely has the highest raw intelligence factor of any model. It is amazing at doing things in 3D, anything involving games, computer use, and subagent coordination. Many benchmarks show dramatic jumps from all previous models. While its performance in coding isn't a quantum leap from Sol, it is very good and makes progress over the previous model. OpenAI has already soft-announced that it has an internal model a level above Astra.

A CACHE HIT IS NOT PROOF THAT YOU SKIPPED THE WORK

A cache hit can be true and still fail to prove that work was skipped. A cache event becomes evidence when the independent oracle expects the prefix, the engine attests it, the prompt path skips it, the output stays identical, the evaluator passes, and the verifier binds those facts to the exact public bundle.

fewer tokens. Stop paying twice for the same answer.

SAKANA: FUGU ULTRA V2

Fugu Ultra v2 is the higher-performance model in Sakana AI's Fugu family. It uses a language model trained to route tasks across a fixed pool of open and specialized models and to recursively call instances of itself. The model prioritizes answer quality on complex multi-step reasoning, autonomous research, and full-stack software development, and does not rely on individual proprietary frontier models in its pool. It supports configurable reasoning effort, function calling, structured outputs, image and PDF input, and built-in web search.

RECURRENT LOOPED TRANSFORMER

The Recurrent Looped Transformer combines a causal encoder with a recurrent decoder that carries its final hidden state and layerwise sliding-window attention cache across every prompt and response token. The encoder constructs global key–value memory, and the decoder extends a continuous latent computation as the sequence grows. The design brings together latent reasoning with unbounded temporal depth, model–hardware co-design, and model–RL algorithm co-design. Realized reasoning gains, hardware efficiency, and RL scaling remain to be established.

SWE BENCHMARK

The new Real-SWE benchmark tests AI models on complex tasks using private enterprise codebases, reflecting real software engineering conditions. With a maximum resolution rate of 38.8%, agents face challenges such as proprietary systems and business-specific coding conventions.

TOOLGRAD: EFFICIENT TOOL-USE DATASET GENERATION WITH TEXTUAL “GRADIENTS”

Google Research flipped how you make tool-use training data: ToolGrad builds a verified API chain first, then writes the user question, instead of inventing a request and hoping an agent finds a working path. That answer-first loop hit a 99.8% success rate on 16,000 real APIs, and a Gemma 3 12B model trained on just 500 of those examples matched Gemini 2.5 Pro on a tool-use test with APIs it had never seen.

HOW GOOD ARE FRONTIER MODELS AT PHYSICS? EXPERT RE-GRADING REVEALS BROKEN EVALUATIONS AND NEAR-SATURATION OF LEADING BENCHMARKS

Those scary physics scores were often the test's fault. Experts rechecked six popular benchmarks and found wrong answer keys, fuzzy questions, and grader bugs behind most model “fails.” Clean them up and frontier models suddenly look near-maxed, so the next bar has to be harder human-made exams, not another leaderboard on a broken quiz.

MANAGED AGENT ARCHITECTURES: WHY FRONTIER LABS ARE REBUILDING THE AGENT LOOP

Frontier labs and cloud providers are turning the agent loop into managed infrastructure, bundling orchestration, versioning, model routing, tools, skills, and optimization behind APIs. Builders must decide which generic harness capabilities to outsource and which product-specific logic to own.

WHO ALIGNS THE ALIGNERS? BRIEF LEGAL THOUGHTS ON THE “AI SAFETY” FIGHTS TO COME

AI brings risks, with some of them, according to some, including the complete destruction of the human race. Some proponents of regulation say that the only appropriate response is total state control. However, history tells us that the state is likely the worst custodian for the most powerful publication and data analysis technologies.

THE FRONTIER NOW SHIPS TWICE. THE SECOND COPY IS NOT FOR SALE.

Anthropic, Google, and OpenAI each shipped their best model twice this month: a public paid tier and a vetted identity-gated tier with the sharper capabilities. Public prices barely moved, but Mythos, Flash Cyber, and Astra's advanced path now ask for org IDs, government ID, or trusted-defender status instead of a bigger budget.

LUXOBENCH

luxobench is a hardware design benchmark for AI models.

CLAUDE FABLE 5.1 SOLVES THE CYPHRAL DISTICH

The Cyphral Distich is a cryptogram consisting of two lines of 32 numbers each that contains a short message deliberately encoded so it can't be read without knowing the rule that produced it.

CHATGPT SITES

ChatGPT Sites now allows users to collaborate and share privately.